摘要详情
46 / 2025-07-18 23:44:26
Dynamic prediction of dam-break flows based on image-driven deep learning model
multiphase fluid dynamics
全文录用
Efficiently predicting dam-break flow is critical for effective flood risk mitigation. This study introduces a deep learning model, Transformer-ResNet-UNet (T-ResUNet), which integrates Residual Network (ResNet), U-shaped Network (UNet), and Transformer architectures to forecast the spatial and temporal dynamics of dam-break water height and flow velocity. Trained on high-fidelity CFD-DEM coupled simulations, T-ResUNet employs recursive input updates for dynamic predictions. Compared to traditional CFD-DEM simulations, T-ResUNet achieves a 1000-fold increase in computational speed while maintaining near-identical accuracy in long-term water wave morphology predictions. Additionally, it significantly outperforms conventional sequence-driven methods in long-term prediction stability. We results demonstrate the superior capability of image-driven deep learning over sequence-based approaches for capturing physical features in dam-break flow predictions.
重要日期
-
会议日期
08月23日
2025
至08月26日
2025
-
07月21日 2025
初稿截稿日期
-
08月26日 2025
注册截止日期
主办单位
Southwest Jiaotong University, China (SWJTU)
International Consortium on Geo-disaster Reduction (ICGdR)
UNESCO Chair on Geoenvironmental Disaster Reduction
International Consortium on Geo-disaster Reduction (ICGdR)
UNESCO Chair on Geoenvironmental Disaster Reduction
承办单位
Southwest Jiaotong University, China (SWJTU)
International Consortium on Geo-disaster Reduction (ICGdR)
UNESCO Chair on Geoenvironmental Disaster Reduction
International Consortium on Geo-disaster Reduction (ICGdR)
UNESCO Chair on Geoenvironmental Disaster Reduction
联系方式
- Yufeng Wang
- wa******@swjtu.edu.cn
- +86********
